Caregiver Speech Mixed Models - 4 categories
Janet Bang
November 17, 2022
This code is for the results to the question: Do features of caregiver verbal engagement vary as a function of language group and activity?
Load libraries and set theme
library(tidyverse)
library(ggpubr)
library(psych)
library(lme4)
library(lmerTest)
library(emmeans)
library(sjPlot)
library(ggeffects)
theme_set(theme_bw())
get_legend<-function(myggplot){
tmp <- ggplot_gtable(ggplot_build(myggplot))
leg <- which(sapply(tmp$grobs, function(x) x$name) == "guide-box")
legend <- tmp$grobs[[leg]]
return(legend)
}Read in data and convert
# demographics
demo_english <- read_csv("./data_demo_lena_transcripts/demo_english_ms.csv") %>%
rename(id = ID, hi = HI24Final, momed = Momed) %>%
dplyr::select(id, hi, momed) %>%
mutate(id = as.character(id),
language = "english")
demo_spanish <- read_csv("./data_demo_lena_transcripts/demo_spanish_ms.csv") %>%
rename(id = ID, hi = HI_18, momed = MomEd_25m) %>%
dplyr::select(id, hi, momed) %>%
mutate(id = as.character(id),
language = "spanish")
demo <- rbind(demo_english, demo_spanish)
# NOTE about periods of non-tCDCS
# gemods refers to when there are designated start/end periods of other-directed speech (ODS); this was captured using gems (@G) using CHAT conventions
# kwalods refers to when ODS was transcribed at an utterance-level within a tCDS activity period between caregiver and child (e.g., other-directed speech in the background); this was captured per utterances using CHAT postcodes
## for tokens/min and types/min, we do not include ODS that occurred within a period of tCDS, because durations were captured by activity and not by utterance
## for mlu, we include all ODS across gemods and kwalods
# NOTE about speech == "all"
# "speech" includes two levels: all, spont
# all = refers to all speech by caregivers
# spont = refers to only speech by caregivers that was considered spontaneous rather than recited (e.g., reading book text, singing memorized common songs like itsy bitsy spider); therefore, 'spont' is a subset of 'all'
# freq
freq <- read_csv("./data_demo_lena_transcripts/freq.csv") %>%
filter(activity != "kwalods",
speech == "all") %>%
mutate(activity = recode(activity, "play" = "cc", "conv" = "cc",
"food" = "cc", "routines" = "cc", "gemods" = "non_tcds")) %>% # we collapsed across play, routines, feeding, and unstructured conversation for the final paper [see supplemental for all categories]
group_by(id, activity, segment_num, speaker) %>%
mutate(dur_min2 = sum(dur_min),
tokens2 = sum(tokens),
types2 = mean(types)) %>%
distinct(id, activity, language, speaker, segment_num, tokens2, types2, dur_min2) %>%
mutate(tokens_permin2 = tokens2/dur_min2,
types_permin2 = types2/dur_min2) %>%
distinct(id, activity, language, speaker, segment_num, tokens_permin2, types_permin2) %>%
ungroup() %>%
mutate(activity = factor(activity, levels = c("books", "cc", "ac", "non_tcds")),
id = factor(id),
language = factor(language))
# mlu
mlu <- read_csv("./data_demo_lena_transcripts/mlu.csv") %>%
filter(speech == "all") %>%
mutate(activity = recode(activity, "play" = "cc", "conv" = "cc",
"food" = "cc", "routines" = "cc", "ods" = "non_tcds")) %>%
group_by(id, activity, segment_num, speaker) %>%
mutate(words_sum2 = sum(words_sum),
num_utt_sum2 = sum(num_utt_sum)) %>%
distinct(id, activity, language, speaker, segment_num, words_sum2, num_utt_sum2) %>%
group_by(id, activity, segment_num, speaker) %>%
mutate(mlu_w2 = words_sum2/num_utt_sum2) %>%
distinct(id, activity, language, speaker, segment_num, mlu_w2) %>%
ungroup() %>%
mutate(activity = factor(activity, levels = c("books", "cc", "ac", "non_tcds")),
id = factor(id),
language = factor(language))
# chip
# this includes only caregivers, therefore there is no speaker column
# we exclude periods of ODS because this is about responsiveness to the child during periods of tCDS
chip <- read_csv("./data_demo_lena_transcripts/chip.csv") %>%
filter(activity != "ods") %>%
mutate(activity = recode(activity, "play" = "cc", "conv" = "cc",
"food" = "cc", "routines" = "cc")) %>%
group_by(id, activity, segment_num) %>%
mutate(total_adult_resp2 = sum(total_adult_resp),
total_adult_imitexp2 = sum(total_adult_imitexp),
total_child_utt2 = sum(total_child_utt)) %>%
distinct(id, activity, language, segment_num, total_adult_resp2, total_adult_imitexp2, total_child_utt2) %>%
mutate(prop_adultresp2 = total_adult_resp2/total_child_utt2,
prop_adult_imitexp2 = total_adult_imitexp2/total_child_utt2) %>%
distinct(id, activity, segment_num, language, prop_adultresp2, prop_adult_imitexp2) %>%
ungroup() %>%
mutate(activity = factor(activity, levels = c("books", "cc", "ac")),
id = factor(id),
language = factor(language))
str(freq)## tibble[,7] [2,870 × 7] (S3: tbl_df/tbl/data.frame)
## $ id : Factor w/ 90 levels "7292","7352",..: 47 47 47 47 50 50 52 52 52 52 ...
## $ activity : Factor w/ 4 levels "books","cc","ac",..: 1 1 1 1 1 1 1 1 1 1 ...
## $ speaker : chr [1:2870] "CHI" "ADULTS" "CHI" "ADULTS" ...
## $ segment_num : num [1:2870] 12 12 15 15 2 2 11 11 5 5 ...
## $ language : Factor w/ 2 levels "english","spanish": 1 1 1 1 1 1 1 1 1 1 ...
## $ tokens_permin2: num [1:2870] 8.46 42.57 5.32 21.75 12.31 ...
## $ types_permin2 : num [1:2870] 4.79 19.73 2.59 9.89 3.61 ...str(mlu)## tibble[,6] [2,594 × 6] (S3: tbl_df/tbl/data.frame)
## $ id : Factor w/ 90 levels "7292","7352",..: 46 46 46 46 46 46 46 46 46 46 ...
## $ activity : Factor w/ 4 levels "books","cc","ac",..: 3 3 2 2 4 4 3 3 2 2 ...
## $ speaker : chr [1:2594] "ADULTS" "CHI" "ADULTS" "CHI" ...
## $ segment_num: num [1:2594] 2 2 2 2 2 2 3 3 3 3 ...
## $ language : Factor w/ 2 levels "english","spanish": 1 1 1 1 1 1 1 1 1 1 ...
## $ mlu_w2 : num [1:2594] 3.18 1.89 2.84 1.73 5.5 ...str(chip)## tibble[,6] [899 × 6] (S3: tbl_df/tbl/data.frame)
## $ activity : Factor w/ 3 levels "books","cc","ac": 3 2 3 2 3 2 2 3 2 3 ...
## $ id : Factor w/ 90 levels "7292","7352",..: 46 46 46 46 46 46 46 46 46 46 ...
## $ language : Factor w/ 2 levels "english","spanish": 1 1 1 1 1 1 1 1 1 1 ...
## $ segment_num : num [1:899] 2 2 3 3 4 4 5 7 7 8 ...
## $ prop_adultresp2 : num [1:899] 1.35 1.57 1.43 1.65 2.14 ...
## $ prop_adult_imitexp2: num [1:899] 0.391 0.418 0.463 0.35 0.643 ...Create dfs for ADULTS and CHI
# FREQ
freq_adult <- freq %>%
filter(speaker == "ADULTS")
# MLU
mlu_adult <- mlu %>%
filter(speaker == "ADULTS")Prep data for mixed models
Create tokens rate per hour - Children
freq_hr_child <- read_csv("./data_demo_lena_transcripts/freq.csv") %>%
dplyr::select(-X1) %>%
filter(speech == "all",
speaker == "CHI") %>%
group_by(id) %>%
mutate(tokens_sum_child = sum(tokens),
dur_hr = sum(dur_min)/60,
tokens_hr_child = tokens_sum_child/dur_hr) %>%
distinct(id, language, tokens_hr_child) %>%
ungroup() %>%
mutate(id = as.character(id))Merge freq_adult, child tokens per hour, and demographic info
# freq
freq_all_mm <- freq_adult %>%
mutate(id = as.character(id)) %>%
left_join(freq_hr_child, by = c("id", "language")) %>%
left_join(demo, by = c("id", "language"))
# mlu
mlu_all_mm <- mlu_adult %>%
mutate(id = as.character(id)) %>%
left_join(freq_hr_child, by = c("id", "language")) %>%
left_join(demo, by = c("id", "language"))
# chip
chip_mm <- chip %>%
mutate(id = as.character(id)) %>%
left_join(freq_hr_child, by = c("id", "language")) %>%
left_join(demo, by = c("id", "language"))Mixed models - 4 categories
TOKENS x ACTIVITY - BOTH - ALL TALK
# # remove outlier - still get interaction
# freq_all_mm2 <- freq_all_mm %>%
# filter(tokens_permin_avg < 530)
# comparing models
# singular fit for MLU with segment_num as random intercept, so removed this for all models
m1_both_tokens_all <- lmer(tokens_permin2 ~ tokens_hr_child + language + activity +
(1 | id),
data = freq_all_mm, REML = F)
m2_both_tokens_all <- lmer(tokens_permin2 ~ tokens_hr_child + (language * activity) +
(1 | id),
data = freq_all_mm, REML = F)
# see if adding the intx adds
anova(m1_both_tokens_all, m2_both_tokens_all)## Data: freq_all_mm
## Models:
## m1_both_tokens_all: tokens_permin2 ~ tokens_hr_child + language + activity + (1 |
## m1_both_tokens_all: id)
## m2_both_tokens_all: tokens_permin2 ~ tokens_hr_child + (language * activity) + (1 |
## m2_both_tokens_all: id)
## npar AIC BIC logLik deviance Chisq Df Pr(>Chisq)
## m1_both_tokens_all 8 14508 14550 -7246.2 14492
## m2_both_tokens_all 11 14504 14562 -7241.3 14482 9.8635 3 0.01976 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# anova table for final model
anova(m2_both_tokens_all)## Type III Analysis of Variance Table with Satterthwaite's method
## Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
## tokens_hr_child 18563 18563 1 89.39 14.1533 0.0003007 ***
## language 19512 19512 1 131.94 14.8765 0.0001787 ***
## activity 333895 111298 3 1379.70 84.8594 < 2.2e-16 ***
## language:activity 12982 4327 3 1379.74 3.2994 0.0197350 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# emmeans
# interaction - post hoc tests of simple effects
m2_both_tokens_all_emmeans_simple1 <- emmeans(m2_both_tokens_all, ~ activity | language)
m2_both_tokens_all_emmeans_simple2 <- emmeans(m2_both_tokens_all, ~ language | activity)
m2_both_tokens_all_emmeans_simple1## language = english:
## activity emmean SE df lower.CL upper.CL
## books 91.6 5.88 1088 80.1 103.1
## cc 66.2 3.18 233 59.9 72.4
## ac 75.5 3.46 315 68.7 82.3
## non_tcds 36.5 3.07 205 30.4 42.5
##
## language = spanish:
## activity emmean SE df lower.CL upper.CL
## books 67.8 6.59 1189 54.8 80.7
## cc 52.9 3.28 261 46.4 59.4
## ac 57.6 3.46 312 50.8 64.4
## non_tcds 31.7 3.07 206 25.6 37.7
##
## Degrees-of-freedom method: kenward-roger
## Confidence level used: 0.95rbind(pairs(m2_both_tokens_all_emmeans_simple1), pairs(m2_both_tokens_all_emmeans_simple2))## language activity contrast estimate SE df t.ratio p.value
## english . books - cc 25.41 5.97 1407 4.260 0.0003
## english . books - ac 16.06 6.17 1419 2.602 0.1499
## english . books - non_tcds 55.11 5.91 1408 9.323 <.0001
## english . cc - ac -9.35 3.62 1376 -2.582 0.1589
## english . cc - non_tcds 29.70 3.24 1357 9.172 <.0001
## english . ac - non_tcds 39.05 3.52 1368 11.103 <.0001
## spanish . books - cc 14.87 6.75 1427 2.203 0.4435
## spanish . books - ac 10.17 6.87 1433 1.480 1.0000
## spanish . books - non_tcds 36.12 6.63 1424 5.450 <.0001
## spanish . cc - ac -4.70 3.71 1382 -1.268 1.0000
## spanish . cc - non_tcds 21.25 3.35 1362 6.351 <.0001
## spanish . ac - non_tcds 25.95 3.52 1370 7.378 <.0001
## . books english - spanish 23.82 8.83 1143 2.697 0.1136
## . cc english - spanish 13.28 4.57 247 2.906 0.0640
## . ac english - spanish 17.93 4.89 312 3.664 0.0047
## . non_tcds english - spanish 4.83 4.35 205 1.112 1.0000
##
## Degrees-of-freedom method: kenward-roger
## P value adjustment: bonferroni method for 16 testsconfint(rbind(pairs(m2_both_tokens_all_emmeans_simple1), pairs(m2_both_tokens_all_emmeans_simple2)))## language activity contrast estimate SE df lower.CL upper.CL
## english . books - cc 25.41 5.97 1407 7.751 43.07
## english . books - ac 16.06 6.17 1419 -2.213 34.33
## english . books - non_tcds 55.11 5.91 1408 37.610 72.61
## english . cc - ac -9.35 3.62 1376 -20.076 1.37
## english . cc - non_tcds 29.70 3.24 1357 20.113 39.28
## english . ac - non_tcds 39.05 3.52 1368 28.639 49.46
## spanish . books - cc 14.87 6.75 1427 -5.108 34.86
## spanish . books - ac 10.17 6.87 1433 -10.175 30.51
## spanish . books - non_tcds 36.12 6.63 1424 16.500 55.74
## spanish . cc - ac -4.70 3.71 1382 -15.687 6.28
## spanish . cc - non_tcds 21.25 3.35 1362 11.343 31.15
## spanish . ac - non_tcds 25.95 3.52 1370 15.538 36.36
## . books english - spanish 23.82 8.83 1143 -2.336 49.97
## . cc english - spanish 13.28 4.57 247 -0.361 26.92
## . ac english - spanish 17.93 4.89 312 3.357 32.50
## . non_tcds english - spanish 4.83 4.35 205 -8.164 17.82
##
## Degrees-of-freedom method: kenward-roger
## Confidence level used: 0.95
## Conf-level adjustment: bonferroni method for 16 estimates# plot
tokens_emmeans <- data.frame(emmeans(m2_both_tokens_all, ~ activity | language))
tokens <- ggplot(tokens_emmeans, aes(activity, emmean, colour = activity, shape = language, group = language)) +
# geom_line(aes(linetype = language)) +
geom_pointrange(aes(ymin = lower.CL, ymax = upper.CL),
position = position_dodge(width = 0.2),
size = 1) +
scale_shape_manual(values = c(16, 21),
name="Language",
labels=c('English', 'Spanish')) +
scale_color_manual(values=c("darkviolet", "firebrick1", "grey", "black"),
name="Activity",
labels=c('Books','Other Child-Cent','Adult-Cent', 'non-tCDS')) +
theme(legend.position = "none",
text = element_text(size = 20),
axis.text.x = element_text(angle = 40, vjust = .9, hjust=.9)) +
scale_x_discrete(labels = c('Books','Other Child-Cent','Adult-Cent', 'non-tCDS')) +
labs(x = "", y = "EMM", title = "Tokens (rate per min)")
tokens
ggsave("./figures/emmeans_tokens_noline.pdf", width = 6, height = 6, units = "in")
# model diagnostics
# only takes into account fixed effects, not random effects
plot_model(m2_both_tokens_all, type = "diag") ## [[1]]
##
## [[2]]
## [[2]]$id
##
##
## [[3]]
##
## [[4]]
TYPES x ACTIVITY - BOTH - ALL TALK
# tried without the outlier and no change
# singular fit for MLU with segment_num as random intercept, so removed this for all models
m1_both_types_all <- lmer(types_permin2 ~ tokens_hr_child + language + activity +
(1 | id),
data = freq_all_mm, REML = F)
m2_both_types_all <- lmer(types_permin2 ~ tokens_hr_child + (language * activity) +
(1 | id),
data = freq_all_mm, REML = F)
# see if adding the intx adds
anova(m1_both_types_all, m2_both_types_all)## Data: freq_all_mm
## Models:
## m1_both_types_all: types_permin2 ~ tokens_hr_child + language + activity + (1 |
## m1_both_types_all: id)
## m2_both_types_all: types_permin2 ~ tokens_hr_child + (language * activity) + (1 |
## m2_both_types_all: id)
## npar AIC BIC logLik deviance Chisq Df Pr(>Chisq)
## m1_both_types_all 8 13902 13944 -6942.9 13886
## m2_both_types_all 11 13898 13956 -6938.2 13876 9.3528 3 0.02495 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# anova table for final model
anova(m2_both_types_all)## Type III Analysis of Variance Table with Satterthwaite's method
## Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
## tokens_hr_child 10217 10217 1 87.37 11.3797 0.001109 **
## language 6063 6063 1 168.75 6.7532 0.010186 *
## activity 160506 53502 3 1392.65 59.5888 < 2.2e-16 ***
## language:activity 8427 2809 3 1392.38 3.1286 0.024896 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# emmeans
# interaction - post hoc tests of simple effects
m2_both_types_all_emmeans_simple1 <- emmeans(m2_both_types_all, ~ activity | language)
m2_both_types_all_emmeans_simple2 <- emmeans(m2_both_types_all, ~ language | activity)
m2_both_types_all_emmeans_simple1## language = english:
## activity emmean SE df lower.CL upper.CL
## books 29.2 4.58 1269 20.2 38.1
## cc 23.9 2.17 406 19.6 28.1
## ac 52.2 2.44 557 47.4 57.0
## non_tcds 20.4 2.07 347 16.3 24.5
##
## language = spanish:
## activity emmean SE df lower.CL upper.CL
## books 23.2 5.16 1283 13.1 33.4
## cc 20.2 2.27 460 15.7 24.6
## ac 38.2 2.44 547 33.5 43.0
## non_tcds 18.5 2.07 350 14.4 22.5
##
## Degrees-of-freedom method: kenward-roger
## Confidence level used: 0.95rbind(pairs(m2_both_types_all_emmeans_simple1), pairs(m2_both_types_all_emmeans_simple2))## language activity contrast estimate SE df t.ratio p.value
## english . books - cc 5.30 4.88 1438 1.086 1.0000
## english . books - ac -23.04 5.03 1444 -4.579 0.0001
## english . books - non_tcds 8.77 4.83 1438 1.813 1.0000
## english . cc - ac -28.34 2.98 1396 -9.499 <.0001
## english . cc - non_tcds 3.47 2.68 1361 1.296 1.0000
## english . ac - non_tcds 31.81 2.90 1383 10.964 <.0001
## spanish . books - cc 3.09 5.49 1443 0.563 1.0000
## spanish . books - ac -15.00 5.58 1438 -2.689 0.1160
## spanish . books - non_tcds 4.79 5.40 1444 0.888 1.0000
## spanish . cc - ac -18.09 3.05 1403 -5.926 <.0001
## spanish . cc - non_tcds 1.70 2.76 1371 0.616 1.0000
## spanish . ac - non_tcds 19.79 2.90 1385 6.823 <.0001
## . books english - spanish 5.92 6.90 1275 0.858 1.0000
## . cc english - spanish 3.71 3.15 433 1.178 1.0000
## . ac english - spanish 13.96 3.45 551 4.040 0.0010
## . non_tcds english - spanish 1.94 2.93 348 0.663 1.0000
##
## Degrees-of-freedom method: kenward-roger
## P value adjustment: bonferroni method for 16 testsconfint(rbind(pairs(m2_both_types_all_emmeans_simple1), pairs(m2_both_types_all_emmeans_simple2)))## language activity contrast estimate SE df lower.CL upper.CL
## english . books - cc 5.30 4.88 1438 -9.15 19.74
## english . books - ac -23.04 5.03 1444 -37.94 -8.14
## english . books - non_tcds 8.77 4.83 1438 -5.54 23.08
## english . cc - ac -28.34 2.98 1396 -37.17 -19.51
## english . cc - non_tcds 3.47 2.68 1361 -4.45 11.39
## english . ac - non_tcds 31.81 2.90 1383 23.22 40.40
## spanish . books - cc 3.09 5.49 1443 -13.16 19.34
## spanish . books - ac -15.00 5.58 1438 -31.51 1.51
## spanish . books - non_tcds 4.79 5.40 1444 -11.19 20.77
## spanish . cc - ac -18.09 3.05 1403 -27.13 -9.05
## spanish . cc - non_tcds 1.70 2.76 1371 -6.48 9.88
## spanish . ac - non_tcds 19.79 2.90 1385 11.20 28.38
## . books english - spanish 5.92 6.90 1275 -14.50 26.34
## . cc english - spanish 3.71 3.15 433 -5.65 13.07
## . ac english - spanish 13.96 3.45 551 3.70 24.21
## . non_tcds english - spanish 1.94 2.93 348 -6.77 10.66
##
## Degrees-of-freedom method: kenward-roger
## Confidence level used: 0.95
## Conf-level adjustment: bonferroni method for 16 estimates# plot
types_emmeans <- data.frame(emmeans(m2_both_types_all, ~ activity | language))
types <- ggplot(types_emmeans, aes(activity, emmean, colour = activity, shape = language, group = language)) +
# geom_line(aes(color = activity)) +
geom_pointrange(aes(ymin = lower.CL, ymax = upper.CL),
position = position_dodge(width = 0.2),
size = 1) +
scale_shape_manual(values = c(16, 21)) +
scale_color_manual(values=c("darkviolet", "firebrick1", "grey", "black")) +
theme(legend.position = "none",
text = element_text(size = 20),
axis.text.x = element_text(angle = 40, vjust = .9, hjust=.9)) +
scale_x_discrete(labels = c('Books','Other Child-Cent','Adult-Cent', 'non-tCDS')) +
labs(x = "", y = "EMM", title = "Types (rate per min)")
types
ggsave("./figures/emmeans_types_noline.pdf", width = 6, height = 6, units = "in")
# model diagnostics
# only takes into account fixed effects, not random effects
plot_model(m2_both_types_all, type = "diag")## [[1]]
##
## [[2]]
## [[2]]$id
##
##
## [[3]]
##
## [[4]]
MLUw x ACTIVITY - BOTH - ALL TALK
# singular fit with segment_num as random intercept, so removed this
m1_both_mlu_all <- lmer(mlu_w2 ~ tokens_hr_child + language + activity +
(1 | id),
data = mlu_all_mm, REML = F)
m2_both_mlu_all <- lmer(mlu_w2 ~ tokens_hr_child + (language * activity) +
(1 | id),
data = mlu_all_mm, REML = F)
# see if adding the intx adds
anova(m1_both_mlu_all, m2_both_mlu_all)## Data: mlu_all_mm
## Models:
## m1_both_mlu_all: mlu_w2 ~ tokens_hr_child + language + activity + (1 | id)
## m2_both_mlu_all: mlu_w2 ~ tokens_hr_child + (language * activity) + (1 | id)
## npar AIC BIC logLik deviance Chisq Df Pr(>Chisq)
## m1_both_mlu_all 8 3683.4 3724.8 -1833.7 3667.4
## m2_both_mlu_all 11 3686.7 3743.6 -1832.3 3664.7 2.6436 3 0.4499# anova table for final model
anova(m1_both_mlu_all)## Type III Analysis of Variance Table with Satterthwaite's method
## Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
## tokens_hr_child 17.496 17.496 1 86.28 19.848 2.502e-05 ***
## language 42.151 42.151 1 88.09 47.818 7.114e-10 ***
## activity 179.377 59.792 3 1251.14 67.831 < 2.2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# emmeans
m1_both_mlu_all_emmeans_activity <- emmeans(m1_both_mlu_all, ~ activity)
m1_both_mlu_all_emmeans_language <- emmeans(m1_both_mlu_all, ~ language)
m1_both_mlu_all_emmeans_activity## activity emmean SE df lower.CL upper.CL
## books 4.30 0.1169 959 4.07 4.53
## cc 3.29 0.0638 200 3.17 3.42
## ac 3.13 0.0682 254 2.99 3.26
## non_tcds 3.90 0.0654 218 3.77 4.03
##
## Results are averaged over the levels of: language
## Degrees-of-freedom method: kenward-roger
## Confidence level used: 0.95m1_both_mlu_all_emmeans_language## language emmean SE df lower.CL upper.CL
## english 4.01 0.0771 106 3.86 4.17
## spanish 3.30 0.0779 110 3.14 3.45
##
## Results are averaged over the levels of: activity
## Degrees-of-freedom method: kenward-roger
## Confidence level used: 0.95pairs(m1_both_mlu_all_emmeans_activity)## contrast estimate SE df t.ratio p.value
## books - cc 1.004 0.1166 1282 8.611 <.0001
## books - ac 1.170 0.1201 1290 9.744 <.0001
## books - non_tcds 0.398 0.1175 1282 3.386 0.0041
## cc - ac 0.166 0.0676 1246 2.451 0.0684
## cc - non_tcds -0.606 0.0647 1239 -9.376 <.0001
## ac - non_tcds -0.772 0.0685 1236 -11.261 <.0001
##
## Results are averaged over the levels of: language
## Degrees-of-freedom method: kenward-roger
## P value adjustment: tukey method for comparing a family of 4 estimates# emmans per language using model WIHTOUT interactions
m1_both_mlu_all_emmeans_simple1 <- emmeans(m1_both_mlu_all, ~ activity | language)
m1_both_mlu_all_emmeans_simple2 <- emmeans(m1_both_mlu_all, ~ language | activity)
m1_both_mlu_all_emmeans_simple1## language = english:
## activity emmean SE df lower.CL upper.CL
## books 4.66 0.1277 583 4.41 4.91
## cc 3.65 0.0824 140 3.49 3.82
## ac 3.49 0.0862 166 3.32 3.66
## non_tcds 4.26 0.0839 150 4.09 4.42
##
## language = spanish:
## activity emmean SE df lower.CL upper.CL
## books 3.94 0.1288 595 3.69 4.19
## cc 2.93 0.0833 145 2.77 3.10
## ac 2.77 0.0863 165 2.60 2.94
## non_tcds 3.54 0.0842 151 3.37 3.71
##
## Degrees-of-freedom method: kenward-roger
## Confidence level used: 0.95rbind(pairs(m1_both_mlu_all_emmeans_simple1), pairs(m1_both_mlu_all_emmeans_simple2))## language activity contrast estimate SE df t.ratio p.value
## english . books - cc 1.004 0.1166 1281.7 8.611 <.0001
## english . books - ac 1.170 0.1201 1289.7 9.744 <.0001
## english . books - non_tcds 0.398 0.1175 1282.1 3.386 0.0117
## english . cc - ac 0.166 0.0676 1246.1 2.451 0.2303
## english . cc - non_tcds -0.606 0.0647 1238.5 -9.376 <.0001
## english . ac - non_tcds -0.772 0.0685 1235.5 -11.261 <.0001
## spanish . books - cc 1.004 0.1166 1281.7 8.611 <.0001
## spanish . books - ac 1.170 0.1201 1289.7 9.744 <.0001
## spanish . books - non_tcds 0.398 0.1175 1282.1 3.386 0.0117
## spanish . cc - ac 0.166 0.0676 1246.1 2.451 0.2303
## spanish . cc - non_tcds -0.606 0.0647 1238.5 -9.376 <.0001
## spanish . ac - non_tcds -0.772 0.0685 1235.5 -11.261 <.0001
## . books english - spanish 0.718 0.1057 93.2 6.794 <.0001
## . cc english - spanish 0.718 0.1057 93.2 6.794 <.0001
## . ac english - spanish 0.718 0.1057 93.2 6.794 <.0001
## . non_tcds english - spanish 0.718 0.1057 93.2 6.794 <.0001
##
## Degrees-of-freedom method: kenward-roger
## P value adjustment: bonferroni method for 16 testsconfint(rbind(pairs(m1_both_mlu_all_emmeans_simple1), pairs(m1_both_mlu_all_emmeans_simple2)))## language activity contrast estimate SE df lower.CL upper.CL
## english . books - cc 1.004 0.1166 1281.7 0.6588 1.349
## english . books - ac 1.170 0.1201 1289.7 0.8144 1.525
## english . books - non_tcds 0.398 0.1175 1282.1 0.0500 0.746
## english . cc - ac 0.166 0.0676 1246.1 -0.0345 0.366
## english . cc - non_tcds -0.606 0.0647 1238.5 -0.7976 -0.415
## english . ac - non_tcds -0.772 0.0685 1235.5 -0.9749 -0.569
## spanish . books - cc 1.004 0.1166 1281.7 0.6588 1.349
## spanish . books - ac 1.170 0.1201 1289.7 0.8144 1.525
## spanish . books - non_tcds 0.398 0.1175 1282.1 0.0500 0.746
## spanish . cc - ac 0.166 0.0676 1246.1 -0.0345 0.366
## spanish . cc - non_tcds -0.606 0.0647 1238.5 -0.7976 -0.415
## spanish . ac - non_tcds -0.772 0.0685 1235.5 -0.9749 -0.569
## . books english - spanish 0.718 0.1057 93.2 0.3973 1.039
## . cc english - spanish 0.718 0.1057 93.2 0.3973 1.039
## . ac english - spanish 0.718 0.1057 93.2 0.3973 1.039
## . non_tcds english - spanish 0.718 0.1057 93.2 0.3973 1.039
##
## Degrees-of-freedom method: kenward-roger
## Confidence level used: 0.95
## Conf-level adjustment: bonferroni method for 16 estimates# plot
mlu_emmeans <- data.frame(emmeans(m1_both_mlu_all, ~ activity | language))
mlu <- ggplot(mlu_emmeans, aes(activity, emmean, colour = activity, shape = language, group = language)) +
# geom_line(aes(color = activity)) +
geom_pointrange(aes(ymin = lower.CL, ymax = upper.CL),
position = position_dodge(width = 0.2),
size = 1) +
scale_shape_manual(values = c(16, 21)) +
scale_color_manual(values=c("darkviolet", "firebrick1", "grey", "black")) +
theme(legend.position = "none",
text = element_text(size = 20),
axis.text.x = element_text(angle = 40, vjust = .9, hjust=.9)) +
scale_x_discrete(labels = c('Books','Other Child-Cent','Adult-Cent', 'non-tCDS')) +
labs(x = "", y = "EMM", title = "MLUw")
ggsave("./figures/emmeans_mlu_noline.pdf", width = 6, height = 6, units = "in")
# model diagnostics
# only takes into account fixed effects, not random effects
plot_model(m1_both_mlu_all, type = "diag")## [[1]]
##
## [[2]]
## [[2]]$id
##
##
## [[3]]
##
## [[4]]
PROP RESPONSES x ACTIVITY - BOTH - ALL TALK
m1_both_propresp_all <- lmer(prop_adultresp2 ~ tokens_hr_child + language + activity +
(1 | id),
data = chip_mm, REML = F)
m2_both_propresp_all <- lmer(prop_adultresp2 ~ tokens_hr_child + (language * activity) +
(1 | id),
data = chip_mm, REML = F)
# see if adding the intx adds
anova(m1_both_propresp_all, m2_both_propresp_all)## Data: chip_mm
## Models:
## m1_both_propresp_all: prop_adultresp2 ~ tokens_hr_child + language + activity + (1 |
## m1_both_propresp_all: id)
## m2_both_propresp_all: prop_adultresp2 ~ tokens_hr_child + (language * activity) + (1 |
## m2_both_propresp_all: id)
## npar AIC BIC logLik deviance Chisq Df Pr(>Chisq)
## m1_both_propresp_all 7 1884.8 1917.8 -935.43 1870.8
## m2_both_propresp_all 9 1882.2 1924.7 -932.12 1864.2 6.6035 2 0.03682 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# anova table for final model
anova(m2_both_propresp_all)## Type III Analysis of Variance Table with Satterthwaite's method
## Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
## tokens_hr_child 12.3461 12.3461 1 83.90 23.7737 5.073e-06 ***
## language 3.1791 3.1791 1 135.49 6.1218 0.01459 *
## activity 18.0184 9.0092 2 801.96 17.3481 4.209e-08 ***
## language:activity 3.4434 1.7217 2 801.53 3.3153 0.03682 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# emmeans
# interaction - post hoc tests of simple effects
m2_both_propresp_all_emmeans_simple1 <- emmeans(m2_both_propresp_all, ~ activity | language)
m2_both_propresp_all_emmeans_simple2 <- emmeans(m2_both_propresp_all, ~ language | activity)
m2_both_propresp_all_emmeans_simple1## language = english:
## activity emmean SE df lower.CL upper.CL
## books 2.44 0.1174 659 2.21 2.67
## cc 1.90 0.0619 168 1.78 2.02
## ac 1.88 0.0727 277 1.74 2.02
##
## language = spanish:
## activity emmean SE df lower.CL upper.CL
## books 2.14 0.1322 690 1.88 2.39
## cc 1.87 0.0644 191 1.74 1.99
## ac 1.59 0.0715 264 1.45 1.73
##
## Degrees-of-freedom method: kenward-roger
## Confidence level used: 0.95rbind(pairs(m2_both_propresp_all_emmeans_simple1), pairs(m2_both_propresp_all_emmeans_simple2))## language activity contrast estimate SE df t.ratio p.value
## english . books - cc 0.5444 0.1204 816 4.522 0.0001
## english . books - ac 0.5623 0.1282 829 4.385 0.0001
## english . cc - ac 0.0180 0.0781 796 0.230 1.0000
## spanish . books - cc 0.2691 0.1367 829 1.969 0.4436
## spanish . books - ac 0.5491 0.1413 830 3.886 0.0010
## spanish . cc - ac 0.2799 0.0788 789 3.551 0.0037
## . books english - spanish 0.3061 0.1767 675 1.732 0.7539
## . cc english - spanish 0.0309 0.0894 180 0.345 1.0000
## . ac english - spanish 0.2929 0.1020 270 2.872 0.0396
##
## Degrees-of-freedom method: kenward-roger
## P value adjustment: bonferroni method for 9 testsconfint(rbind(pairs(m2_both_propresp_all_emmeans_simple1), pairs(m2_both_propresp_all_emmeans_simple2)))## language activity contrast estimate SE df lower.CL upper.CL
## english . books - cc 0.5444 0.1204 816 0.20963 0.879
## english . books - ac 0.5623 0.1282 829 0.20577 0.919
## english . cc - ac 0.0180 0.0781 796 -0.19930 0.235
## spanish . books - cc 0.2691 0.1367 829 -0.11088 0.649
## spanish . books - ac 0.5491 0.1413 830 0.15622 0.942
## spanish . cc - ac 0.2799 0.0788 789 0.06073 0.499
## . books english - spanish 0.3061 0.1767 675 -0.18559 0.798
## . cc english - spanish 0.0309 0.0894 180 -0.22005 0.282
## . ac english - spanish 0.2929 0.1020 270 0.00782 0.578
##
## Degrees-of-freedom method: kenward-roger
## Confidence level used: 0.95
## Conf-level adjustment: bonferroni method for 9 estimates# plot
propresp_emmeans <- data.frame(emmeans(m2_both_propresp_all, ~ activity | language))
resp <- ggplot(propresp_emmeans, aes(activity, emmean, colour = activity, shape = language, group = language)) +
# geom_line(aes(color = activity)) +
geom_pointrange(aes(ymin = lower.CL, ymax = upper.CL),
position = position_dodge(width = 0.2),
size = 1) +
scale_shape_manual(values = c(16, 21)) +
scale_color_manual(values=c("darkviolet", "firebrick1", "grey", "black")) +
theme(legend.position = "none",
text = element_text(size = 20),
axis.text.x = element_text(angle = 40, vjust = .9, hjust=.9)) +
scale_x_discrete(labels = c('Books','Other Child-Cent','Adult-Cent')) +
labs(x = "", y = "EMM", title = "Proportion of Responses")
ggsave("./figures/emmeans_resp_noline.pdf", width = 6, height = 6, units = "in")
# model diagnostics
# only takes into account fixed effects, not random effects
plot_model(m1_both_propresp_all, type = "diag")## [[1]]
##
## [[2]]
## [[2]]$id
##
##
## [[3]]
##
## [[4]]
PROP IMIT/EXP x ACTIVITY - BOTH - ALL TALK
# # tried removing two outliers, but no change in findings
# chip_mm2 <- chip_mm %>%
# filter(prop_adult_imitexp_avg < 2.9)
m1_both_propimitexp_all <- lmer(prop_adult_imitexp2 ~ tokens_hr_child + language + activity +
(1 | id),
data = chip_mm, REML = F)
m2_both_propimitexp_all <- lmer(prop_adult_imitexp2 ~ tokens_hr_child + (language * activity) +
(1 | id),
data = chip_mm, REML = F)
# see if adding the intx adds
anova(m1_both_propimitexp_all, m2_both_propimitexp_all)## Data: chip_mm
## Models:
## m1_both_propimitexp_all: prop_adult_imitexp2 ~ tokens_hr_child + language + activity +
## m1_both_propimitexp_all: (1 | id)
## m2_both_propimitexp_all: prop_adult_imitexp2 ~ tokens_hr_child + (language * activity) +
## m2_both_propimitexp_all: (1 | id)
## npar AIC BIC logLik deviance Chisq Df Pr(>Chisq)
## m1_both_propimitexp_all 7 456.64 489.63 -221.32 442.64
## m2_both_propimitexp_all 9 458.71 501.12 -220.35 440.71 1.9303 2 0.3809# anova table for final model
anova(m1_both_propimitexp_all)## Type III Analysis of Variance Table with Satterthwaite's method
## Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
## tokens_hr_child 0.13640 0.13640 1 86.42 1.4590 0.230382
## language 0.11961 0.11961 1 86.58 1.2794 0.261126
## activity 0.98600 0.49300 2 804.92 5.2736 0.005304 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# emmeans
m1_both_propimitexp_all_emmeans_activity <- emmeans(m1_both_propimitexp_all, ~ activity)
m1_both_propimitexp_all_emmeans_language <- emmeans(m1_both_propimitexp_all, ~ language)
m1_both_propimitexp_all_emmeans_activity## activity emmean SE df lower.CL upper.CL
## books 0.448 0.0366 683 0.376 0.520
## cc 0.390 0.0181 189 0.354 0.426
## ac 0.333 0.0209 289 0.292 0.374
##
## Results are averaged over the levels of: language
## Degrees-of-freedom method: kenward-roger
## Confidence level used: 0.95m1_both_propimitexp_all_emmeans_language## language emmean SE df lower.CL upper.CL
## english 0.407 0.0227 118 0.362 0.452
## spanish 0.374 0.0233 127 0.328 0.420
##
## Results are averaged over the levels of: activity
## Degrees-of-freedom method: kenward-roger
## Confidence level used: 0.95pairs(m1_both_propimitexp_all_emmeans_activity)## contrast estimate SE df t.ratio p.value
## books - cc 0.0581 0.0381 825 1.526 0.2793
## books - ac 0.1151 0.0400 828 2.880 0.0114
## cc - ac 0.0569 0.0234 795 2.428 0.0407
##
## Results are averaged over the levels of: language
## Degrees-of-freedom method: kenward-roger
## P value adjustment: tukey method for comparing a family of 3 estimates# emmans per language using model WIHTOUT interactions
m1_both_propimitexp_all_emmeans_simple1 <- emmeans(m1_both_propimitexp_all, ~ activity | language)
m1_both_propimitexp_all_emmeans_simple2 <- emmeans(m1_both_propimitexp_all, ~ language | activity)
m1_both_propimitexp_all_emmeans_simple1## language = english:
## activity emmean SE df lower.CL upper.CL
## books 0.465 0.0393 500 0.388 0.542
## cc 0.407 0.0232 132 0.361 0.453
## ac 0.350 0.0259 189 0.299 0.401
##
## language = spanish:
## activity emmean SE df lower.CL upper.CL
## books 0.431 0.0400 514 0.353 0.510
## cc 0.373 0.0239 144 0.326 0.421
## ac 0.316 0.0257 183 0.266 0.367
##
## Degrees-of-freedom method: kenward-roger
## Confidence level used: 0.95rbind(pairs(m1_both_propimitexp_all_emmeans_simple1), pairs(m1_both_propimitexp_all_emmeans_simple2))## language activity contrast estimate SE df t.ratio p.value
## english . books - cc 0.0581 0.0381 824.7 1.526 1.0000
## english . books - ac 0.1151 0.0400 828.0 2.880 0.0367
## english . cc - ac 0.0569 0.0234 795.3 2.428 0.1385
## spanish . books - cc 0.0581 0.0381 824.7 1.526 1.0000
## spanish . books - ac 0.1151 0.0400 828.0 2.880 0.0367
## spanish . cc - ac 0.0569 0.0234 795.3 2.428 0.1385
## . books english - spanish 0.0335 0.0301 92.7 1.110 1.0000
## . cc english - spanish 0.0335 0.0301 92.7 1.110 1.0000
## . ac english - spanish 0.0335 0.0301 92.7 1.110 1.0000
##
## Degrees-of-freedom method: kenward-roger
## P value adjustment: bonferroni method for 9 testsconfint(rbind(pairs(m1_both_propimitexp_all_emmeans_simple1), pairs(m1_both_propimitexp_all_emmeans_simple2)))## language activity contrast estimate SE df lower.CL upper.CL
## english . books - cc 0.0581 0.0381 824.7 -0.04780 0.164
## english . books - ac 0.1151 0.0400 828.0 0.00399 0.226
## english . cc - ac 0.0569 0.0234 795.3 -0.00826 0.122
## spanish . books - cc 0.0581 0.0381 824.7 -0.04780 0.164
## spanish . books - ac 0.1151 0.0400 828.0 0.00399 0.226
## spanish . cc - ac 0.0569 0.0234 795.3 -0.00826 0.122
## . books english - spanish 0.0335 0.0301 92.7 -0.05213 0.119
## . cc english - spanish 0.0335 0.0301 92.7 -0.05213 0.119
## . ac english - spanish 0.0335 0.0301 92.7 -0.05213 0.119
##
## Degrees-of-freedom method: kenward-roger
## Confidence level used: 0.95
## Conf-level adjustment: bonferroni method for 9 estimates# plot
propimitexp_emmeans <- data.frame(emmeans(m1_both_propimitexp_all, ~ activity | language))
imit_exp <- ggplot(propimitexp_emmeans, aes(activity, emmean, colour = activity, shape = language, group = language)) +
# geom_line(aes(color = activity)) +
geom_pointrange(aes(ymin = lower.CL, ymax = upper.CL),
position = position_dodge(width = 0.2),
size = 1) +
scale_shape_manual(values = c(16, 21)) +
scale_color_manual(values=c("darkviolet", "firebrick1", "grey", "black")) +
theme(legend.position = "none",
text = element_text(size = 20),
axis.text.x = element_text(angle = 40, vjust = .9, hjust=.9)) +
scale_x_discrete(labels = c('Books','Other Child-Cent','Adult-Cent')) +
labs(x = "", y = "EMM", title = "Proportion of Imitations/Expansions")
ggsave("./figures/emmeans_imit_exp_noline.pdf", width = 6, height = 6, units = "in")
# model diagnostics
# only takes into account fixed effects, not random effects
plot_model(m1_both_propimitexp_all, type = "diag")## [[1]]
##
## [[2]]
## [[2]]$id
##
##
## [[3]]
##
## [[4]]
Combining plots (NO LINE)
# grid all
ggarrange(tokens, types, mlu, resp, imit_exp, common.legend = T, legend = "bottom")
ggsave("./figures/emmeans_grid_all_noline.pdf", width = 14, height = 10, units = "in", dpi = 300)Checking HI as Covariate
TOKENS x ACTIVITY - BOTH - ALL TALK - WITH HI
# comparing models
# singular fit for MLU with segment_num as random intercept, so removed this for all models
m1_both_tokens_all_hi <- lmer(tokens_permin2 ~ tokens_hr_child + language + activity +
(1 | id),
data = freq_all_mm, REML = F)
m2_both_tokens_all_hi <- lmer(tokens_permin2 ~ tokens_hr_child + (language * activity) +
(1 | id),
data = freq_all_mm, REML = F)
m3_both_tokens_all_hi <- lmer(tokens_permin2 ~ tokens_hr_child + (language * activity) + hi +
(1 | id),
data = freq_all_mm, REML = F)
# see if adding mom ed after the intx adds
anova(m1_both_tokens_all_hi, m2_both_tokens_all_hi, m3_both_tokens_all_hi)## Data: freq_all_mm
## Models:
## m1_both_tokens_all_hi: tokens_permin2 ~ tokens_hr_child + language + activity + (1 |
## m1_both_tokens_all_hi: id)
## m2_both_tokens_all_hi: tokens_permin2 ~ tokens_hr_child + (language * activity) + (1 |
## m2_both_tokens_all_hi: id)
## m3_both_tokens_all_hi: tokens_permin2 ~ tokens_hr_child + (language * activity) + hi +
## m3_both_tokens_all_hi: (1 | id)
## npar AIC BIC logLik deviance Chisq Df Pr(>Chisq)
## m1_both_tokens_all_hi 8 14508 14550 -7246.2 14492
## m2_both_tokens_all_hi 11 14504 14562 -7241.3 14482 9.8635 3 0.01976 *
## m3_both_tokens_all_hi 12 14506 14570 -7241.2 14482 0.1263 1 0.72228
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# anova table for model with intx
anova(m3_both_tokens_all_hi)## Type III Analysis of Variance Table with Satterthwaite's method
## Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
## tokens_hr_child 18658 18658 1 89.40 14.2257 0.0002908 ***
## language 17357 17357 1 123.81 13.2333 0.0004021 ***
## activity 334012 111337 3 1379.16 84.8874 < 2.2e-16 ***
## hi 166 166 1 89.36 0.1264 0.7229869
## language:activity 12987 4329 3 1379.30 3.3006 0.0197043 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1TYPES x ACTIVITY - BOTH - ALL TALK - WITH HI
# tried without the outlier and no change
# singular fit for MLU with segment_num as random intercept, so removed this for all models
m1_both_types_all_hi <- lmer(types_permin2 ~ tokens_hr_child + language + activity +
(1 | id),
data = freq_all_mm, REML = F)
m2_both_types_all_hi <- lmer(types_permin2 ~ tokens_hr_child + (language * activity) +
(1 | id),
data = freq_all_mm, REML = F)
m3_both_types_all_hi <- lmer(types_permin2 ~ tokens_hr_child + (language * activity) + hi +
(1 | id),
data = freq_all_mm, REML = F)
# see if adding mom ed after the intx adds
anova(m1_both_types_all_hi, m2_both_types_all_hi, m3_both_types_all_hi)## Data: freq_all_mm
## Models:
## m1_both_types_all_hi: types_permin2 ~ tokens_hr_child + language + activity + (1 |
## m1_both_types_all_hi: id)
## m2_both_types_all_hi: types_permin2 ~ tokens_hr_child + (language * activity) + (1 |
## m2_both_types_all_hi: id)
## m3_both_types_all_hi: types_permin2 ~ tokens_hr_child + (language * activity) + hi +
## m3_both_types_all_hi: (1 | id)
## npar AIC BIC logLik deviance Chisq Df Pr(>Chisq)
## m1_both_types_all_hi 8 13902 13944 -6942.9 13886
## m2_both_types_all_hi 11 13898 13956 -6938.2 13876 9.3528 3 0.02495 *
## m3_both_types_all_hi 12 13900 13964 -6938.2 13876 0.0363 1 0.84898
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# anova table for model with intx
anova(m3_both_types_all_hi)## Type III Analysis of Variance Table with Satterthwaite's method
## Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
## tokens_hr_child 9738 9738 1 87.52 10.8466 0.001429 **
## language 4729 4729 1 153.03 5.2672 0.023089 *
## activity 160531 53510 3 1392.87 59.6000 < 2.2e-16 ***
## hi 33 33 1 87.72 0.0363 0.849423
## language:activity 8447 2816 3 1392.18 3.1360 0.024647 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1MLUw x ACTIVITY - BOTH - ALL TALK - WITH HI
# singular fit with segment_num as random intercept, so removed this
m1_both_mlu_all_hi <- lmer(mlu_w2 ~ tokens_hr_child + language + activity +
(1 | id),
data = mlu_all_mm, REML = F)
m2_both_mlu_all_hi <- lmer(mlu_w2 ~ tokens_hr_child + (language * activity) +
(1 | id),
data = mlu_all_mm, REML = F)
m3_both_mlu_all_hi <- lmer(mlu_w2 ~ tokens_hr_child + (language * activity) + hi +
(1 | id),
data = mlu_all_mm, REML = F)
# see if adding mom ed after the intx adds
anova(m1_both_mlu_all_hi, m2_both_mlu_all_hi, m3_both_mlu_all_hi)## Data: mlu_all_mm
## Models:
## m1_both_mlu_all_hi: mlu_w2 ~ tokens_hr_child + language + activity + (1 | id)
## m2_both_mlu_all_hi: mlu_w2 ~ tokens_hr_child + (language * activity) + (1 | id)
## m3_both_mlu_all_hi: mlu_w2 ~ tokens_hr_child + (language * activity) + hi + (1 |
## m3_both_mlu_all_hi: id)
## npar AIC BIC logLik deviance Chisq Df Pr(>Chisq)
## m1_both_mlu_all_hi 8 3683.4 3724.8 -1833.7 3667.4
## m2_both_mlu_all_hi 11 3686.7 3743.6 -1832.3 3664.7 2.6436 3 0.4499
## m3_both_mlu_all_hi 12 3688.4 3750.5 -1832.2 3664.4 0.2936 1 0.5879# anova table for model with intx
anova(m3_both_mlu_all_hi)## Type III Analysis of Variance Table with Satterthwaite's method
## Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
## tokens_hr_child 16.606 16.606 1 86.33 18.8694 3.80e-05 ***
## language 32.530 32.530 1 111.85 36.9639 1.71e-08 ***
## activity 175.612 58.537 3 1251.43 66.5171 < 2.2e-16 ***
## hi 0.259 0.259 1 87.31 0.2941 0.5890
## language:activity 2.304 0.768 3 1251.71 0.8727 0.4546
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1PROP RESPONSES x ACTIVITY - BOTH - ALL TALK - WITH HI
m1_both_propresp_all_hi <- lmer(prop_adultresp2 ~ tokens_hr_child + language + activity +
(1 | id),
data = chip_mm, REML = F)
m2_both_propresp_all_hi <- lmer(prop_adultresp2 ~ tokens_hr_child + (language * activity) +
(1 | id),
data = chip_mm, REML = F)
m3_both_propresp_all_hi <- lmer(prop_adultresp2 ~ tokens_hr_child + (language * activity) + hi +
(1 | id),
data = chip_mm, REML = F)
# see if adding mom ed after the intx adds
anova(m1_both_propresp_all_hi, m2_both_propresp_all_hi, m3_both_propresp_all_hi)## Data: chip_mm
## Models:
## m1_both_propresp_all_hi: prop_adultresp2 ~ tokens_hr_child + language + activity + (1 |
## m1_both_propresp_all_hi: id)
## m2_both_propresp_all_hi: prop_adultresp2 ~ tokens_hr_child + (language * activity) + (1 |
## m2_both_propresp_all_hi: id)
## m3_both_propresp_all_hi: prop_adultresp2 ~ tokens_hr_child + (language * activity) + hi +
## m3_both_propresp_all_hi: (1 | id)
## npar AIC BIC logLik deviance Chisq Df Pr(>Chisq)
## m1_both_propresp_all_hi 7 1884.8 1917.8 -935.43 1870.8
## m2_both_propresp_all_hi 9 1882.2 1924.7 -932.12 1864.2 6.6035 2 0.03682 *
## m3_both_propresp_all_hi 10 1883.3 1930.4 -931.64 1863.3 0.9754 1 0.32333
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# anova table for model with intx
anova(m3_both_propresp_all_hi)## Type III Analysis of Variance Table with Satterthwaite's method
## Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
## tokens_hr_child 11.4097 11.4097 1 83.05 21.9660 1.074e-05 ***
## language 3.7108 3.7108 1 125.12 7.1440 0.008524 **
## activity 18.3616 9.1808 2 801.92 17.6749 3.078e-08 ***
## hi 0.5101 0.5101 1 81.40 0.9821 0.324620
## language:activity 3.4075 1.7037 2 801.05 3.2801 0.038132 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1PROP IMIT/EXP x ACTIVITY - BOTH - ALL TALK - WITH HI
m1_both_propimitexp_all_hi <- lmer(prop_adult_imitexp2 ~ tokens_hr_child + language + activity +
(1 | id),
data = chip_mm, REML = F)
m2_both_propimitexp_all_hi <- lmer(prop_adult_imitexp2 ~ tokens_hr_child + (language * activity) +
(1 | id),
data = chip_mm, REML = F)
m3_both_propimitexp_all_hi <- lmer(prop_adult_imitexp2 ~ tokens_hr_child + (language * activity) + hi +
(1 | id),
data = chip_mm, REML = F)
# see if adding mom ed after the intx adds
anova(m1_both_propimitexp_all_hi, m2_both_propimitexp_all_hi, m3_both_propimitexp_all_hi)## Data: chip_mm
## Models:
## m1_both_propimitexp_all_hi: prop_adult_imitexp2 ~ tokens_hr_child + language + activity +
## m1_both_propimitexp_all_hi: (1 | id)
## m2_both_propimitexp_all_hi: prop_adult_imitexp2 ~ tokens_hr_child + (language * activity) +
## m2_both_propimitexp_all_hi: (1 | id)
## m3_both_propimitexp_all_hi: prop_adult_imitexp2 ~ tokens_hr_child + (language * activity) +
## m3_both_propimitexp_all_hi: hi + (1 | id)
## npar AIC BIC logLik deviance Chisq Df Pr(>Chisq)
## m1_both_propimitexp_all_hi 7 456.64 489.63 -221.32 442.64
## m2_both_propimitexp_all_hi 9 458.71 501.12 -220.35 440.71 1.9303 2 0.3809
## m3_both_propimitexp_all_hi 10 459.03 506.15 -219.51 439.03 1.6806 1 0.1948# anova table for model with intx
anova(m3_both_propimitexp_all_hi)## Type III Analysis of Variance Table with Satterthwaite's method
## Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
## tokens_hr_child 0.17684 0.17684 1 85.85 1.8957 0.172139
## language 0.09446 0.09446 1 133.68 1.0126 0.316109
## activity 0.89726 0.44863 2 805.51 4.8091 0.008391 **
## hi 0.15810 0.15810 1 84.07 1.6948 0.196525
## language:activity 0.16891 0.08445 2 804.34 0.9053 0.404827
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1